Starlink Satellite Detection Pipeline

Challenge

Manual satellite signal detection was consuming 8+ hours per week across the team. The existing system could only handle a fraction of the satellite constellation and had inconsistent accuracy across different signal types.

Solution

Built an ML pipeline using signal processing and deep learning to automatically detect satellite signals. Implemented feature extraction from raw signal data, trained a CNN model on historical observations, and created a real-time detection system. Integrated with existing tracking infrastructure and built monitoring dashboards.

Key Learnings

• •Signal processing requires domain expertise—worked closely with RF engineers to understand noise characteristics
• •Model generalization is critical when dealing with real-world signals across seasons and weather
• •Monitoring and alerting are as important as the model itself—false negatives caused bigger issues than false positives
• •Starting with a simple baseline and iterating beats trying to build the perfect system from day one

Technologies Used